Tetsuji Haga

Robust abandoned object detection using dual foregrounds

As an alternative to the tracking-based approaches that heavily depend on accurate detection of moving objects, which often fail for crowded scenarios, we present a pixelwise method that employs dual foregrounds to extract temporally static image regions. Depending on the application, these regions indicate objects that do not constitute the original background but were brought into the scene at a subsequent time, such as abandoned and removed items, illegally parked vehicles. We construct separate long- and short-term backgrounds that are implemented as pixelwise multivariate Gaussian models. Background parameters are adapted online using a Bayesian update mechanism imposed at different learning rates. By comparing each frame with these models, we estimate two foregrounds. We infer an evidence score at each pixel by applying a set of hypotheses on the foreground responses, and then aggregate the evidence in time to provide temporal consistency. Unlike optical flow-based approaches that smear boundaries, our method can accurately segment out objects even if they are fully occluded. It does not require on-site training to compensate for particular imaging conditions. While having a low-computational load, it readily lends itself to parallelization if further speed improvement is necessary.

Human detection in outdoor scene using spatio-temporal motion analysis

We propose an image processing algorithm for detecting human in outdoor scenes containing changeful background. In this work, regions extracted through background subtraction procedure are accurately classified into human and others by motion analysis in the three dimensional feature space constructed by the spatial uniqueness of image motion F/sub 1/, the temporal uniqueness of image motion F/sub 2/, and the temporal motion continuity F/sub 3/. Evaluation test proved that proposed algorithm could reduce the error rates of both false positive and false negative to about 1/3 compared with a conventional method. We also tested it in a PC-based real-time system over two weeks in real environments, that resulted in its false negative error rate of less than 1% and false positive error number of less than 3 times per day.

A fast algorithm of video super-resolution using dimensionality reduction by DCT and example selection

In this paper, we propose a novel learning-based video super resolution algorithm with less memory requirements and computational cost. To this end, we adopt discrete cosine transform (DCT) coefficients for feature vector components. Moreover, we design an example selection procedure to construct a compact database. We conducted evaluative experiments using MPEG test sequences to synthesize a high resolution video. Experimental results show that our method can improve effectiveness of super-resolution algorithm, while preserving the quality of synthesized image.

Robust Change-Detection by Normalised Gradient-Correlation

A novel algorithm for robustly segmenting changes between different images of a scene is presented. This computationally efficient algorithm is based on a non-linear comparison of gradient structure in overlapping image-regions and offers intrinsic invariance to changing illumination, without recourse to background-model adaptation. High accuracy is demonstrated on test video data with and without illumination changes. The technique is applicable to motion-segmentation as well as measuring longer-term object-changes.

Human specific activity retrieval from a surveillance image sequence

We propose an image processing system which searches the moving human and vehicles from the long-term surveillance video for intruder detection or parking lot monitoring, by comparing a series of retrieval queries like passing a certain area, moving direction, duration time and so on. In such a system, not only on-line detection from real-time video by pre-defined query but also quick re-search going back to the past when the user changed the query, and post-retrieval by interactive query change are required. In proposed method, we simplified the image retrieval meta-data to be described in a small and fixed length data however complicated the trajectory of moving object is. Moreover, the matching core function of image retrieval process is realized by a simple comparator. That enables fast image retrieval however complicated the series of the retrieval queries is.

Construction Method of Efficient Database for Learning-Based Video Super-Resolution

There are two major problems with learning-based super-resolution algorithms. One is that they require a large amount of memory to store examples; while the other is the high computational cost of finding the nearest neighbors in the database. In order to alleviate these problems, it is helpful to reduce the dimensionality of examples and to store only a small number of examples that contribute to the synthesis of a high quality video. Based on these ideas, we have developed an efficient algorithm for learning-based video super-resolution. We introduce several strategies to construct an efficient database. Through the evaluation experiments we show the efficiency of our approach in improving super-resolution algorithms.

Vision system for object handling robot using a low-resolution range image and an intensity image

In this paper, we propose an object recognition system which consists of two modules. One is an object extraction module using a low-resolution range image. The other is a position measurement module for detecting precise position of each object using an intensity edge image. To acquire a range image, we propose stereo vision with random-dot pattern projection. This method enables reliable stereo matching, even if there is no texture on the objects. The equipment is simple because a variable pattern is not necessary. We found this system has 99.8% recognition reliability and the processing time is approximately 5 s/image, providing satisfactory performance for practical use.

Depth-based Selective Image Reconstruction Using Spatiotemporal Image Analysis

In industrial plants, a remote monitoring system which removes physical tour inspection is often considered desirable. However the image sequence given from the mobile inspection robot is hard to see because interested objects are often partially occluded by obstacles such as pillars or fences. Our aim is to improve the image sequence that increases the efficiency and reliability of remote visual inspection. We propose a new depth-based image processing technique, which removes the needless objects from the foreground and recovers the occluded background electronically. Our algorithm is based on spatiotemporal analysis that enables fine and dense depth estimation, depth-based precise segmentation, and accurate interpolation. We apply this technique to a real time sequence given from the mobile inspection robot. The resulted image sequence is satisfactory in that the operator can make correct visual inspection with less fatigue.